The Spectrum & Clinical Tools for Reliable Assessment (SPECTRA) Evidence Stack for Adult Neurodivergence Assessment: Framework Development and Synthetic Data Evaluation
Background: Psychological assessment of adult neurodivergence relies on multi-informant data that frequently disagree; discrepancies may reflect masking, context-dependent expression, or limited self-insight rather than measurement error. Objective: This paper specifies the SPECTRA (Spectrum & Clinical Tools for Reliable Assessment) Evidence Stack, a rule-based, neuroaffirmative architecture that preserves cross-informant discrepancies as interpretable evidence rather than collapsing them into opaque scores, producing quality flags, convergence profiles, evidence-strength bands, and reason codes. Methods: This is a design specification, not a validation study: no clinical participants were involved. Framework behavior was evaluated on 250 author-generated synthetic records through a 4-condition, prespecified rule-configuration comparison and a threshold-perturbation sensitivity analysis. Stochastic generation and score jitter were controlled by a recorded random seed. Results: With the disclosed code, version, and seed, the reference run was reproducible: mean evidence strength was 0.60 for the full-stack configuration, 0.53 for the no-quality comparator, 0.54 for the no-convergence comparator, and 0.50 for the baseline comparator, with the ordering A > C > B > D preserved in all 12 threshold-perturbation scenarios. These descriptive differences verify the behavior of the 4 specified configurations; they are not causal estimates of the isolated effects of individual layers. An extended sensitivity analysis covering all 22 reference-configuration parameters (one-at-a-time ±10%, a 1000-sample Latin hypercube joint perturbation of ±20%, and tipping-point sweeps of the band cutoffs) showed that the ordering was preserved in 38 of 44 single-parameter scenarios but not under joint perturbation (278/1000, 27.8% of samples); the caution cap and the high-band cutoff were the dominant drivers of band assignment. The reference configuration is fully disclosed and independently runnable. Conclusions: Within the disclosed reference configuration, the SPECTRA Evidence Stack produced reproducible, conditionally assigned evidence-strength outputs that preserved cross-informant discrepancy as interpretable evidence and made uncertainty visible through bands, flags, and reason codes; the ordering of the 4 rule configurations was preserved in most single-parameter scenarios but not under joint perturbation, and the parameters most influential for band assignment (the caution cap and the high-band cutoff) were identified for priority calibration. Diagnostic accuracy, criterion validity, reliability, and clinical utility are not evaluated here and are reserved for a subsequent clinical validation study, for which a prespecified calibration protocol is now provided.